On 2009-10-13 at 15:04:36 [+0800], Axel Dörfler <axeld@xxxxxxxxxxxxxxxx> wrote: > "Michael Lotz" <mmlr@xxxxxxxx> wrote: > > > Maybe they can one day be made more light-weight and faster in > > > userland, too. > > The rw_lock uses the thread blocking API through the corresponding > > syscall though, so it shouldn't be too bad. > > Good! Indeed. Just if you consider working some more on those, I had that idea about introducing a similar mechanism currently used for preventing signal delivery for preventing a thread from being preempted for a very short time. That would allow for cheap userland spinlocks, which could also be used to improve the userland locking primitives (I was originally thinking of the pthread ones). > > > All in all, I think I would prefer a nice C++ API for userland for > > > these things, as that is our primary API. > > We can certainly come up with a satisfying solution at BG :-) > > That, and so many other things :-) Sounds like a very full schedule. :-) CU, Ingo